xref: /openbmc/linux/fs/nfsd/nfs4layouts.c (revision 01cfbad7)
1 /*
2  * Copyright (c) 2014 Christoph Hellwig.
3  */
4 #include <linux/kmod.h>
5 #include <linux/file.h>
6 #include <linux/jhash.h>
7 #include <linux/sched.h>
8 #include <linux/sunrpc/addr.h>
9 
10 #include "pnfs.h"
11 #include "netns.h"
12 #include "trace.h"
13 
14 #define NFSDDBG_FACILITY                NFSDDBG_PNFS
15 
16 struct nfs4_layout {
17 	struct list_head		lo_perstate;
18 	struct nfs4_layout_stateid	*lo_state;
19 	struct nfsd4_layout_seg		lo_seg;
20 };
21 
22 static struct kmem_cache *nfs4_layout_cache;
23 static struct kmem_cache *nfs4_layout_stateid_cache;
24 
25 static const struct nfsd4_callback_ops nfsd4_cb_layout_ops;
26 static const struct lock_manager_operations nfsd4_layouts_lm_ops;
27 
28 const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] =  {
29 	[LAYOUT_BLOCK_VOLUME]	= &bl_layout_ops,
30 };
31 
32 /* pNFS device ID to export fsid mapping */
33 #define DEVID_HASH_BITS	8
34 #define DEVID_HASH_SIZE	(1 << DEVID_HASH_BITS)
35 #define DEVID_HASH_MASK	(DEVID_HASH_SIZE - 1)
36 static u64 nfsd_devid_seq = 1;
37 static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE];
38 static DEFINE_SPINLOCK(nfsd_devid_lock);
39 
40 static inline u32 devid_hashfn(u64 idx)
41 {
42 	return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK;
43 }
44 
45 static void
46 nfsd4_alloc_devid_map(const struct svc_fh *fhp)
47 {
48 	const struct knfsd_fh *fh = &fhp->fh_handle;
49 	size_t fsid_len = key_len(fh->fh_fsid_type);
50 	struct nfsd4_deviceid_map *map, *old;
51 	int i;
52 
53 	map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL);
54 	if (!map)
55 		return;
56 
57 	map->fsid_type = fh->fh_fsid_type;
58 	memcpy(&map->fsid, fh->fh_fsid, fsid_len);
59 
60 	spin_lock(&nfsd_devid_lock);
61 	if (fhp->fh_export->ex_devid_map)
62 		goto out_unlock;
63 
64 	for (i = 0; i < DEVID_HASH_SIZE; i++) {
65 		list_for_each_entry(old, &nfsd_devid_hash[i], hash) {
66 			if (old->fsid_type != fh->fh_fsid_type)
67 				continue;
68 			if (memcmp(old->fsid, fh->fh_fsid,
69 					key_len(old->fsid_type)))
70 				continue;
71 
72 			fhp->fh_export->ex_devid_map = old;
73 			goto out_unlock;
74 		}
75 	}
76 
77 	map->idx = nfsd_devid_seq++;
78 	list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]);
79 	fhp->fh_export->ex_devid_map = map;
80 	map = NULL;
81 
82 out_unlock:
83 	spin_unlock(&nfsd_devid_lock);
84 	kfree(map);
85 }
86 
87 struct nfsd4_deviceid_map *
88 nfsd4_find_devid_map(int idx)
89 {
90 	struct nfsd4_deviceid_map *map, *ret = NULL;
91 
92 	rcu_read_lock();
93 	list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash)
94 		if (map->idx == idx)
95 			ret = map;
96 	rcu_read_unlock();
97 
98 	return ret;
99 }
100 
101 int
102 nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp,
103 		u32 device_generation)
104 {
105 	if (!fhp->fh_export->ex_devid_map) {
106 		nfsd4_alloc_devid_map(fhp);
107 		if (!fhp->fh_export->ex_devid_map)
108 			return -ENOMEM;
109 	}
110 
111 	id->fsid_idx = fhp->fh_export->ex_devid_map->idx;
112 	id->generation = device_generation;
113 	id->pad = 0;
114 	return 0;
115 }
116 
117 void nfsd4_setup_layout_type(struct svc_export *exp)
118 {
119 	struct super_block *sb = exp->ex_path.mnt->mnt_sb;
120 
121 	if (!(exp->ex_flags & NFSEXP_PNFS))
122 		return;
123 
124 	if (sb->s_export_op->get_uuid &&
125 	    sb->s_export_op->map_blocks &&
126 	    sb->s_export_op->commit_blocks)
127 		exp->ex_layout_type = LAYOUT_BLOCK_VOLUME;
128 }
129 
130 static void
131 nfsd4_free_layout_stateid(struct nfs4_stid *stid)
132 {
133 	struct nfs4_layout_stateid *ls = layoutstateid(stid);
134 	struct nfs4_client *clp = ls->ls_stid.sc_client;
135 	struct nfs4_file *fp = ls->ls_stid.sc_file;
136 
137 	trace_layoutstate_free(&ls->ls_stid.sc_stateid);
138 
139 	spin_lock(&clp->cl_lock);
140 	list_del_init(&ls->ls_perclnt);
141 	spin_unlock(&clp->cl_lock);
142 
143 	spin_lock(&fp->fi_lock);
144 	list_del_init(&ls->ls_perfile);
145 	spin_unlock(&fp->fi_lock);
146 
147 	vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls);
148 	fput(ls->ls_file);
149 
150 	if (ls->ls_recalled)
151 		atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
152 
153 	kmem_cache_free(nfs4_layout_stateid_cache, ls);
154 }
155 
156 static int
157 nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
158 {
159 	struct file_lock *fl;
160 	int status;
161 
162 	fl = locks_alloc_lock();
163 	if (!fl)
164 		return -ENOMEM;
165 	locks_init_lock(fl);
166 	fl->fl_lmops = &nfsd4_layouts_lm_ops;
167 	fl->fl_flags = FL_LAYOUT;
168 	fl->fl_type = F_RDLCK;
169 	fl->fl_end = OFFSET_MAX;
170 	fl->fl_owner = ls;
171 	fl->fl_pid = current->tgid;
172 	fl->fl_file = ls->ls_file;
173 
174 	status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL);
175 	if (status) {
176 		locks_free_lock(fl);
177 		return status;
178 	}
179 	BUG_ON(fl != NULL);
180 	return 0;
181 }
182 
183 static struct nfs4_layout_stateid *
184 nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
185 		struct nfs4_stid *parent, u32 layout_type)
186 {
187 	struct nfs4_client *clp = cstate->clp;
188 	struct nfs4_file *fp = parent->sc_file;
189 	struct nfs4_layout_stateid *ls;
190 	struct nfs4_stid *stp;
191 
192 	stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache);
193 	if (!stp)
194 		return NULL;
195 	stp->sc_free = nfsd4_free_layout_stateid;
196 	get_nfs4_file(fp);
197 	stp->sc_file = fp;
198 
199 	ls = layoutstateid(stp);
200 	INIT_LIST_HEAD(&ls->ls_perclnt);
201 	INIT_LIST_HEAD(&ls->ls_perfile);
202 	spin_lock_init(&ls->ls_lock);
203 	INIT_LIST_HEAD(&ls->ls_layouts);
204 	mutex_init(&ls->ls_mutex);
205 	ls->ls_layout_type = layout_type;
206 	nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
207 			NFSPROC4_CLNT_CB_LAYOUT);
208 
209 	if (parent->sc_type == NFS4_DELEG_STID)
210 		ls->ls_file = get_file(fp->fi_deleg_file);
211 	else
212 		ls->ls_file = find_any_file(fp);
213 	BUG_ON(!ls->ls_file);
214 
215 	if (nfsd4_layout_setlease(ls)) {
216 		fput(ls->ls_file);
217 		put_nfs4_file(fp);
218 		kmem_cache_free(nfs4_layout_stateid_cache, ls);
219 		return NULL;
220 	}
221 
222 	spin_lock(&clp->cl_lock);
223 	stp->sc_type = NFS4_LAYOUT_STID;
224 	list_add(&ls->ls_perclnt, &clp->cl_lo_states);
225 	spin_unlock(&clp->cl_lock);
226 
227 	spin_lock(&fp->fi_lock);
228 	list_add(&ls->ls_perfile, &fp->fi_lo_states);
229 	spin_unlock(&fp->fi_lock);
230 
231 	trace_layoutstate_alloc(&ls->ls_stid.sc_stateid);
232 	return ls;
233 }
234 
235 __be32
236 nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
237 		struct nfsd4_compound_state *cstate, stateid_t *stateid,
238 		bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
239 {
240 	struct nfs4_layout_stateid *ls;
241 	struct nfs4_stid *stid;
242 	unsigned char typemask = NFS4_LAYOUT_STID;
243 	__be32 status;
244 
245 	if (create)
246 		typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
247 
248 	status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
249 			net_generic(SVC_NET(rqstp), nfsd_net_id));
250 	if (status)
251 		goto out;
252 
253 	if (!fh_match(&cstate->current_fh.fh_handle,
254 		      &stid->sc_file->fi_fhandle)) {
255 		status = nfserr_bad_stateid;
256 		goto out_put_stid;
257 	}
258 
259 	if (stid->sc_type != NFS4_LAYOUT_STID) {
260 		ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
261 		nfs4_put_stid(stid);
262 
263 		status = nfserr_jukebox;
264 		if (!ls)
265 			goto out;
266 		mutex_lock(&ls->ls_mutex);
267 	} else {
268 		ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
269 
270 		status = nfserr_bad_stateid;
271 		mutex_lock(&ls->ls_mutex);
272 		if (stateid->si_generation > stid->sc_stateid.si_generation)
273 			goto out_unlock_stid;
274 		if (layout_type != ls->ls_layout_type)
275 			goto out_unlock_stid;
276 	}
277 
278 	*lsp = ls;
279 	return 0;
280 
281 out_unlock_stid:
282 	mutex_unlock(&ls->ls_mutex);
283 out_put_stid:
284 	nfs4_put_stid(stid);
285 out:
286 	return status;
287 }
288 
289 static void
290 nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
291 {
292 	spin_lock(&ls->ls_lock);
293 	if (ls->ls_recalled)
294 		goto out_unlock;
295 
296 	ls->ls_recalled = true;
297 	atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
298 	if (list_empty(&ls->ls_layouts))
299 		goto out_unlock;
300 
301 	trace_layout_recall(&ls->ls_stid.sc_stateid);
302 
303 	atomic_inc(&ls->ls_stid.sc_count);
304 	nfsd4_run_cb(&ls->ls_recall);
305 
306 out_unlock:
307 	spin_unlock(&ls->ls_lock);
308 }
309 
310 static inline u64
311 layout_end(struct nfsd4_layout_seg *seg)
312 {
313 	u64 end = seg->offset + seg->length;
314 	return end >= seg->offset ? end : NFS4_MAX_UINT64;
315 }
316 
317 static void
318 layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
319 {
320 	if (end == NFS4_MAX_UINT64)
321 		lo->length = NFS4_MAX_UINT64;
322 	else
323 		lo->length = end - lo->offset;
324 }
325 
326 static bool
327 layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
328 {
329 	if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
330 		return false;
331 	if (layout_end(&lo->lo_seg) <= s->offset)
332 		return false;
333 	if (layout_end(s) <= lo->lo_seg.offset)
334 		return false;
335 	return true;
336 }
337 
338 static bool
339 layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
340 {
341 	if (lo->iomode != new->iomode)
342 		return false;
343 	if (layout_end(new) < lo->offset)
344 		return false;
345 	if (layout_end(lo) < new->offset)
346 		return false;
347 
348 	lo->offset = min(lo->offset, new->offset);
349 	layout_update_len(lo, max(layout_end(lo), layout_end(new)));
350 	return true;
351 }
352 
353 static __be32
354 nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
355 {
356 	struct nfs4_file *fp = ls->ls_stid.sc_file;
357 	struct nfs4_layout_stateid *l, *n;
358 	__be32 nfserr = nfs_ok;
359 
360 	assert_spin_locked(&fp->fi_lock);
361 
362 	list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
363 		if (l != ls) {
364 			nfsd4_recall_file_layout(l);
365 			nfserr = nfserr_recallconflict;
366 		}
367 	}
368 
369 	return nfserr;
370 }
371 
372 __be32
373 nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
374 {
375 	struct nfsd4_layout_seg *seg = &lgp->lg_seg;
376 	struct nfs4_file *fp = ls->ls_stid.sc_file;
377 	struct nfs4_layout *lp, *new = NULL;
378 	__be32 nfserr;
379 
380 	spin_lock(&fp->fi_lock);
381 	nfserr = nfsd4_recall_conflict(ls);
382 	if (nfserr)
383 		goto out;
384 	spin_lock(&ls->ls_lock);
385 	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
386 		if (layouts_try_merge(&lp->lo_seg, seg))
387 			goto done;
388 	}
389 	spin_unlock(&ls->ls_lock);
390 	spin_unlock(&fp->fi_lock);
391 
392 	new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
393 	if (!new)
394 		return nfserr_jukebox;
395 	memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg));
396 	new->lo_state = ls;
397 
398 	spin_lock(&fp->fi_lock);
399 	nfserr = nfsd4_recall_conflict(ls);
400 	if (nfserr)
401 		goto out;
402 	spin_lock(&ls->ls_lock);
403 	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
404 		if (layouts_try_merge(&lp->lo_seg, seg))
405 			goto done;
406 	}
407 
408 	atomic_inc(&ls->ls_stid.sc_count);
409 	list_add_tail(&new->lo_perstate, &ls->ls_layouts);
410 	new = NULL;
411 done:
412 	nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid);
413 	spin_unlock(&ls->ls_lock);
414 out:
415 	spin_unlock(&fp->fi_lock);
416 	if (new)
417 		kmem_cache_free(nfs4_layout_cache, new);
418 	return nfserr;
419 }
420 
421 static void
422 nfsd4_free_layouts(struct list_head *reaplist)
423 {
424 	while (!list_empty(reaplist)) {
425 		struct nfs4_layout *lp = list_first_entry(reaplist,
426 				struct nfs4_layout, lo_perstate);
427 
428 		list_del(&lp->lo_perstate);
429 		nfs4_put_stid(&lp->lo_state->ls_stid);
430 		kmem_cache_free(nfs4_layout_cache, lp);
431 	}
432 }
433 
434 static void
435 nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
436 		struct list_head *reaplist)
437 {
438 	struct nfsd4_layout_seg *lo = &lp->lo_seg;
439 	u64 end = layout_end(lo);
440 
441 	if (seg->offset <= lo->offset) {
442 		if (layout_end(seg) >= end) {
443 			list_move_tail(&lp->lo_perstate, reaplist);
444 			return;
445 		}
446 		lo->offset = layout_end(seg);
447 	} else {
448 		/* retain the whole layout segment on a split. */
449 		if (layout_end(seg) < end) {
450 			dprintk("%s: split not supported\n", __func__);
451 			return;
452 		}
453 		end = seg->offset;
454 	}
455 
456 	layout_update_len(lo, end);
457 }
458 
459 __be32
460 nfsd4_return_file_layouts(struct svc_rqst *rqstp,
461 		struct nfsd4_compound_state *cstate,
462 		struct nfsd4_layoutreturn *lrp)
463 {
464 	struct nfs4_layout_stateid *ls;
465 	struct nfs4_layout *lp, *n;
466 	LIST_HEAD(reaplist);
467 	__be32 nfserr;
468 	int found = 0;
469 
470 	nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
471 						false, lrp->lr_layout_type,
472 						&ls);
473 	if (nfserr) {
474 		trace_layout_return_lookup_fail(&lrp->lr_sid);
475 		return nfserr;
476 	}
477 
478 	spin_lock(&ls->ls_lock);
479 	list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
480 		if (layouts_overlapping(lp, &lrp->lr_seg)) {
481 			nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
482 			found++;
483 		}
484 	}
485 	if (!list_empty(&ls->ls_layouts)) {
486 		if (found)
487 			nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid);
488 		lrp->lrs_present = 1;
489 	} else {
490 		trace_layoutstate_unhash(&ls->ls_stid.sc_stateid);
491 		nfs4_unhash_stid(&ls->ls_stid);
492 		lrp->lrs_present = 0;
493 	}
494 	spin_unlock(&ls->ls_lock);
495 
496 	mutex_unlock(&ls->ls_mutex);
497 	nfs4_put_stid(&ls->ls_stid);
498 	nfsd4_free_layouts(&reaplist);
499 	return nfs_ok;
500 }
501 
502 __be32
503 nfsd4_return_client_layouts(struct svc_rqst *rqstp,
504 		struct nfsd4_compound_state *cstate,
505 		struct nfsd4_layoutreturn *lrp)
506 {
507 	struct nfs4_layout_stateid *ls, *n;
508 	struct nfs4_client *clp = cstate->clp;
509 	struct nfs4_layout *lp, *t;
510 	LIST_HEAD(reaplist);
511 
512 	lrp->lrs_present = 0;
513 
514 	spin_lock(&clp->cl_lock);
515 	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
516 		if (ls->ls_layout_type != lrp->lr_layout_type)
517 			continue;
518 
519 		if (lrp->lr_return_type == RETURN_FSID &&
520 		    !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
521 				   &cstate->current_fh.fh_handle))
522 			continue;
523 
524 		spin_lock(&ls->ls_lock);
525 		list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
526 			if (lrp->lr_seg.iomode == IOMODE_ANY ||
527 			    lrp->lr_seg.iomode == lp->lo_seg.iomode)
528 				list_move_tail(&lp->lo_perstate, &reaplist);
529 		}
530 		spin_unlock(&ls->ls_lock);
531 	}
532 	spin_unlock(&clp->cl_lock);
533 
534 	nfsd4_free_layouts(&reaplist);
535 	return 0;
536 }
537 
538 static void
539 nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
540 		struct list_head *reaplist)
541 {
542 	spin_lock(&ls->ls_lock);
543 	list_splice_init(&ls->ls_layouts, reaplist);
544 	spin_unlock(&ls->ls_lock);
545 }
546 
547 void
548 nfsd4_return_all_client_layouts(struct nfs4_client *clp)
549 {
550 	struct nfs4_layout_stateid *ls, *n;
551 	LIST_HEAD(reaplist);
552 
553 	spin_lock(&clp->cl_lock);
554 	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
555 		nfsd4_return_all_layouts(ls, &reaplist);
556 	spin_unlock(&clp->cl_lock);
557 
558 	nfsd4_free_layouts(&reaplist);
559 }
560 
561 void
562 nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
563 {
564 	struct nfs4_layout_stateid *ls, *n;
565 	LIST_HEAD(reaplist);
566 
567 	spin_lock(&fp->fi_lock);
568 	list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
569 		if (ls->ls_stid.sc_client == clp)
570 			nfsd4_return_all_layouts(ls, &reaplist);
571 	}
572 	spin_unlock(&fp->fi_lock);
573 
574 	nfsd4_free_layouts(&reaplist);
575 }
576 
577 static void
578 nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
579 {
580 	struct nfs4_client *clp = ls->ls_stid.sc_client;
581 	char addr_str[INET6_ADDRSTRLEN];
582 	static char *envp[] = {
583 		"HOME=/",
584 		"TERM=linux",
585 		"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
586 		NULL
587 	};
588 	char *argv[8];
589 	int error;
590 
591 	rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
592 
593 	trace_layout_recall_fail(&ls->ls_stid.sc_stateid);
594 
595 	printk(KERN_WARNING
596 		"nfsd: client %s failed to respond to layout recall. "
597 		"  Fencing..\n", addr_str);
598 
599 	argv[0] = "/sbin/nfsd-recall-failed";
600 	argv[1] = addr_str;
601 	argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id;
602 	argv[3] = NULL;
603 
604 	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
605 	if (error) {
606 		printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
607 			addr_str, error);
608 	}
609 }
610 
611 static void
612 nfsd4_cb_layout_prepare(struct nfsd4_callback *cb)
613 {
614 	struct nfs4_layout_stateid *ls =
615 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
616 
617 	mutex_lock(&ls->ls_mutex);
618 	nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid);
619 	mutex_unlock(&ls->ls_mutex);
620 }
621 
622 static int
623 nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
624 {
625 	struct nfs4_layout_stateid *ls =
626 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
627 	struct nfsd_net *nn;
628 	ktime_t now, cutoff;
629 	LIST_HEAD(reaplist);
630 
631 
632 	switch (task->tk_status) {
633 	case 0:
634 	case -NFS4ERR_DELAY:
635 		/*
636 		 * Anything left? If not, then call it done. Note that we don't
637 		 * take the spinlock since this is an optimization and nothing
638 		 * should get added until the cb counter goes to zero.
639 		 */
640 		if (list_empty(&ls->ls_layouts))
641 			return 1;
642 
643 		/* Poll the client until it's done with the layout */
644 		now = ktime_get();
645 		nn = net_generic(ls->ls_stid.sc_client->net, nfsd_net_id);
646 
647 		/* Client gets 2 lease periods to return it */
648 		cutoff = ktime_add_ns(task->tk_start,
649 					 nn->nfsd4_lease * NSEC_PER_SEC * 2);
650 
651 		if (ktime_before(now, cutoff)) {
652 			rpc_delay(task, HZ/100); /* 10 mili-seconds */
653 			return 0;
654 		}
655 		/* Fallthrough */
656 	case -NFS4ERR_NOMATCHING_LAYOUT:
657 		trace_layout_recall_done(&ls->ls_stid.sc_stateid);
658 		task->tk_status = 0;
659 		return 1;
660 	default:
661 		/*
662 		 * Unknown error or non-responding client, we'll need to fence.
663 		 */
664 		nfsd4_cb_layout_fail(ls);
665 		return -1;
666 	}
667 }
668 
669 static void
670 nfsd4_cb_layout_release(struct nfsd4_callback *cb)
671 {
672 	struct nfs4_layout_stateid *ls =
673 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
674 	LIST_HEAD(reaplist);
675 
676 	trace_layout_recall_release(&ls->ls_stid.sc_stateid);
677 
678 	nfsd4_return_all_layouts(ls, &reaplist);
679 	nfsd4_free_layouts(&reaplist);
680 	nfs4_put_stid(&ls->ls_stid);
681 }
682 
683 static const struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
684 	.prepare	= nfsd4_cb_layout_prepare,
685 	.done		= nfsd4_cb_layout_done,
686 	.release	= nfsd4_cb_layout_release,
687 };
688 
689 static bool
690 nfsd4_layout_lm_break(struct file_lock *fl)
691 {
692 	/*
693 	 * We don't want the locks code to timeout the lease for us;
694 	 * we'll remove it ourself if a layout isn't returned
695 	 * in time:
696 	 */
697 	fl->fl_break_time = 0;
698 	nfsd4_recall_file_layout(fl->fl_owner);
699 	return false;
700 }
701 
702 static int
703 nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
704 		struct list_head *dispose)
705 {
706 	BUG_ON(!(arg & F_UNLCK));
707 	return lease_modify(onlist, arg, dispose);
708 }
709 
710 static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
711 	.lm_break	= nfsd4_layout_lm_break,
712 	.lm_change	= nfsd4_layout_lm_change,
713 };
714 
715 int
716 nfsd4_init_pnfs(void)
717 {
718 	int i;
719 
720 	for (i = 0; i < DEVID_HASH_SIZE; i++)
721 		INIT_LIST_HEAD(&nfsd_devid_hash[i]);
722 
723 	nfs4_layout_cache = kmem_cache_create("nfs4_layout",
724 			sizeof(struct nfs4_layout), 0, 0, NULL);
725 	if (!nfs4_layout_cache)
726 		return -ENOMEM;
727 
728 	nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
729 			sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
730 	if (!nfs4_layout_stateid_cache) {
731 		kmem_cache_destroy(nfs4_layout_cache);
732 		return -ENOMEM;
733 	}
734 	return 0;
735 }
736 
737 void
738 nfsd4_exit_pnfs(void)
739 {
740 	int i;
741 
742 	kmem_cache_destroy(nfs4_layout_cache);
743 	kmem_cache_destroy(nfs4_layout_stateid_cache);
744 
745 	for (i = 0; i < DEVID_HASH_SIZE; i++) {
746 		struct nfsd4_deviceid_map *map, *n;
747 
748 		list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
749 			kfree(map);
750 	}
751 }
752